Researcher-$\in -the$-Loop Evolutionary Discovery of Entanglement-Assisted Quantum Code Families

AI involvement AI-centered

The researchers supplied targets, a seed program, construction directions and an exact evaluator. AlphaEvolve revised search programs and produced the finite qubit constructions before the family formula was given in prompts. Researchers then analyzed the output and supplied the generalization and proof. This supports AI-centered discovery.

• Baek, Won and Kim: Researcher-$\in -the$-Loop Evolutionary Discovery of Entanglement-Assisted Quantum Code Families — Introduction and Sections 3.2–3.4

Importance Notable

The evolved finite codes improve specific archived distance bounds, and their pattern led researchers to a broader proved family. The 54 table-gap closures combine the human-derived family with entanglement lifting; they are not all direct outputs of the model.

Technical verification Formal (Magma exact finite-field checks)

The paper reports that Magma $V2.28-20$ independently recomputed code parameters for 115 core codes and 33 family instances with zero mismatches. It also reports Lean and Comparator checks of the researcher-derived family theorem. These are source-reported machine checks; this assessment selects the finite AI-produced constructions.

• Code matrices, saved programs and verification artifact

Human understanding Explained

Baek, Won and Kim explain the evolved search, show a representative program, identify the pattern in the finite codes, and derive the general family themselves. Their account explains the central construction and its route to the theorem.